ClusterPointIndicesDecomposer

What is this

Decompose jsk_recognition_msgs/ClusterPointIndices into array of topics of sensor_msgs/PointCloud like ~output00, ~output01 and so on. It also publishes tf of centroids of each cluster and oriented bounding box of them. The direction of the bounding box are aligned on to the nearest planes if available.

Subscribing topics

• ~input (sensor_msgs/PointCloud2):

  Input pointcloud.

• ~target (jsk_recognition_msgs/ClusterPointIndices):

  Input set of indices to represent clusters.

• ~align_planes (jsk_recognition_msgs/PolygonArray):

• ~align_planes_coefficients (jsk_recognition_msgs/ModelCoefficientsArray):

  The planes for bounding box to be aligned on.

Publishing topics

Default Topics

• ~debug_output (sensor_msgs/PointCloud2):

  Concatenate all the clusters into one pointcloud and colorize each cluster to see the result of segmentation.

• ~boxes (jsk_recognition_msgs/BoundingBoxArray):

  Array of oriented bounding box for each segmented cluster.

  If ~align_boxes, ~align_boxes_with_plane and ~fill_boxes_label_with_nearest_plane_index are True, each box(jsk_recognition_msgs/BoundingBox)’s label indicates nearest plane index.

• ~label (sensor_msgs/Image):

  Label image for each cluster point indices. You can visualize it with jsk_perception/ColorizeLabels

• ~mask (sensor_msgs/Image):

  Mask image generated from cluster point indices.

• ~centroid_pose_array (geometry_msgs/PoseArray)

  Each cloud’s centroid poses.

• ~negative_indices (pcl_msgs/PointIndices)

  Point indices which are not included in input indices.

• ~cluster_indices (jsk_recognition_msgs/ClusterPointIndices)

  Sorted cluster point indices.

Optional Topics

• ~output%02d (sensor_msgs/PointCloud2):

  Series of topics for each pointcloud cluster. This is published only when publish_clouds is true.

• /tf (tf2_msgs/TFMessage)

  Transformation of each decomposed points’ center of gravity.

Parameters

• ~approximate_sync (Boolean, default: False):

  Policy of synchronization, if False it synchronizes exactly, else approximately.

• ~queue_size (Int, default: 100):

  Queue size of topic msgs for synchronization.

• ~publish_tf (Boolean, default: False):

  Toggle tf publishing.

• ~publish_clouds (Boolean, default: False):

  Toggle ~output%02d topics.

• ~align_boxes (Boolean, default: False):

  • ~align_boxes_with_plane (Boolean, default: True):

    • Is enabled only if ~align_boxes is True.

  • If ~align_boxes is True and ~align_boxes_with_plane is True:

    • Topics ~align_planes and ~align_planes_coefficients are enabled.

  • If ~align_boxes is True and ~align_boxes_with_plane is False:

    • Parameter ~target_frame_id is required, and bounding boxes are aligned with the target frame.

  See sample/sample_cluster_point_indices_decomposer.launch also.

• ~use_pca (Boolean, default: False):

  Run PCA algorithm on each cluster to estimate x and y direction. The x-axis indicates the first principal component and y-axis indicates the second one.

• ~force_to_flip_z_axis (Boolean, default: True)

  Flip z axis direction if this value is true.

• ~max_size, ~min_size (Int, default: -1, -1)

  If positive value, ignores cluster which points size is external from (clusterPointsSize < ~min_size or clusterPointsSize > ~max_size).

• ~sort_by (String, default input_indices)

  Sort output indices. Currently below options are supported:]

  • input_indices: same order as the input cluster indices

  • z_axis: sort by z axis of cloud

  • cloud_size: sort by size of cloud

• ~fill_boxes_label_with_nearest_plane_index (Boolean, default: False):

  If ~align_boxes, ~align_boxes_with_plane and this value are True, each box(jsk_recognition_msgs/BoundingBox)’s label indicates nearest plane index.

Sample

roslaunch jsk_pcl_ros sample_cluster_point_indices_decomposer.launch
roslaunch jsk_pcl_ros sample_cluster_point_indices_decomposer_sort_by.launch